IDENT: Improving design and analysis of oncology trials Evaluating new Targeted Therapies

Background:
Phase II clinical trials are often designed to assess whether a new therapy works in patients on average. In cancer trials, however, patients can respond very differently to the same treatment. This could be because of differences in tumour mutations. Modern ‘targeted treatments’ are developed to target particular genetic make-up of the tumour instead of the location of the cancer in the body. A new trial approach called ‘basket trials’ can enrol patients of various cancer types, for example, lung cancer and breast cancer patients that share similar genetic profiles. Current approaches often
1) analyse the subtrials, e.g., different cancer types, separately,
2) require an unnecessarily large number of patients to establish efficacy, and
3) do not permit making changes as the basket trial continues.

Aims:
This fellowship aims to investigate how basket trials can be best designed and analysed. Specifically, the research will propose statistical methodology that can: 1) fully utilise the subtrial data, allowing borrowing of information, to improve decision making,
2) plan basket trials with a smaller sample size, which may be re-estimated as the trial continues,
3) prioritise development paths in certain patient subgroups using adaptive designs.

Methods:
I will use real cancer studies such as the MAJIC trial to motivate the methodological research for basket trials with added efficiency. Bayesian approaches will be developed to permit information from relevant subtrials to be represented into a prior for the treatment effect. The magnitude of borrowing will depend on how consistent the treatment’s effect is. Formulae to calculate sample size for basket trials will be derived, incorporating a new parameter for subtrial data consistency. The level of data consistency will first be assumed as a known quantity, informed by e.g., biological knowledge, to obtain a fixed sample size when standard non-adaptive tests are used. Adaptive designs will also be developed for basket trials to permit re-estimating the sample size as data on the consistency is gathered, in order to ensure the study power is correct.

How the results of this research will be used:
I will publish papers on statistical methods and their applications. Open-source software with user-friendly web interfaces will be released. The methods will allow a substantial gain in power and estimation accuracy for basket trials and better value for funders of trials such as CRUK. In the long term, methods will allow more efficient clinical evaluation and better treatments for cancer patients.

Clinical trials are used to test the risk and benefit of a new therapy before it can be made widely available. These trials have been designed to test if the ‘average’ patient benefits. However, in trials of new cancer treatments, patients can respond very differently to the same treatment. This is often because tumours have different mutations. Modern ‘targeted treatments’ are developed to target particular genetic make-up of the tumour instead of the location of the cancer in the body. My work will focus on a new trial approach called ‘basket trials’, which can enrol patients of different cancer types, for example, lung cancer and breast cancer that share similar genetic profiles, to receive the same anti-cancer therapy. There are two main advantages of basket trials. Firstly, it is easier, quicker and cheaper to run a basket trial than assess each cancer location separately. Secondly, basket trials allow patient information to be fully used for understanding how the therapy works in different cancer types. Current approaches often
1) analyse each part of a basket trial separately,
2) ask for a maximum number of patients to establish the efficacy, which is inefficient and exposes more patients to an unproven treatment than might be necessary, and
3) do not allow making changes as the trial continues.

My statistical expertise will help improve basket trials to find effective cancer treatments. I will develop analysis models to make better use of basket trial data. They will allow sharing of information between similar cancer types. This will mean that we get a better understanding about which treatments work, and for whom. These new statistical approaches will make a difference to the way we design basket trials. I will propose new methods to calculate the required number of patients to recruit in a basket trial. Because information from similar parts can be combined, a considerably smaller number of patients will be needed. More flexible designs will also be developed to allow making changes as the basket trial continues. For example, it would be possible to stop some parts earlier than planned, if evidence shows a meaningful improvement in patients’ outcome. As a result, basket trials can be completed quicker and at a lower cost, allowing new therapies to go from creation to use in practice considerably quicker. In this way, cancer patients can be assured to be treated much quicker and with more effective therapies.